G4 CUT&Tag analysis mESC (WT, FANCJ KO, DHX36 KO, DKO)

Simon Elsässer, Karolinska Institutet (2023)

DESeq2 analysis of peak universe (joint peak set called in any condition, must be called in all three replicates, MACS3)

bw_dir <- "/Volumes/DATA/DATA/Puck/bigwig/"
chromhmm <- "../genome/ESC_10_segments.mm39.bed"

library("wigglescout")
library("ggpubr")
library("ggplot2")
library("DESeq2")
library("dplyr")
library("ggrastr")

clean <- function (fn) {
  fn <- gsub(pattern = ".+/", "", x = fn)
  fn <- gsub(pattern = ".mm9.+", "", x = fn)
  fn <- gsub(pattern = ".mm39.+", "", x = fn)
  fn <- gsub(pattern = "_S.+", "", x = fn)
  fn <- gsub(pattern = ".scaled.bw", "", x = fn)
  fn <- gsub(pattern = ".unscaled.bw", "", x = fn)
  fn <- gsub(pattern = "_batch2", "", x = fn)
  fn <- gsub(pattern = "-", " ", x = fn)
  fn <- gsub(pattern = "_", " ", x = fn)
  fn <- gsub(pattern = " HA ", " ", x = fn)
  fn <- gsub(pattern = "D1D6", "FANCJ-/-", x = fn)
  fn <- gsub(pattern = "P2D2", "DHX36-/-", x = fn)
  fn <- gsub(pattern = "P3D4", "FANCJ-/-DHX36-/-", x = fn)
  return(fn)
}

BWs <- paste0(bw_dir,list.files(bw_dir,pattern="G4_.+R.\\.bw"))

mypal <-c("cornflowerblue","orange","red2","#505050")
mypal2 <- rep(mypal,each=2)
mypal3 <-c("cornflowerblue","cornflowerblue","cornflowerblue","orange","orange","orange","red2","red2","red2","505050","505050","505050")

bw_granges_diff_analysis <- function(granges_c1,
                                     granges_c2,
                                     label_c1,
                                     label_c2,
                                     estimate_size_factors = FALSE,
                                     as_granges = FALSE) {

  # Bind first, get numbers after
  names_values <- NULL
  fields <- names(mcols(granges_c1))

  if ("name" %in% fields) {
    names_values <- mcols(granges_c1)[["name"]]
    granges_c1 <- granges_c1[, fields[fields != "name"]]
  }

  fields <- names(mcols(granges_c2))
  if ("name" %in% fields) {
    granges_c2 <- granges_c2[, fields[fields != "name"]]
  }

  cts_df <- cbind(data.frame(granges_c1), mcols(granges_c2))

  if (! is.null(names_values)) {
    rownames(cts_df) <- names_values
  }

  # Needs to drop non-complete cases and match rows
  complete <- complete.cases(cts_df)
  cts_df <- cts_df[complete, ]

  values_df <- cts_df[, 6:ncol(cts_df)] %>% dplyr::select(where(is.numeric))
  cts <- get_nreads_columns(values_df)

  condition_labels <- c(rep(label_c1, length(mcols(granges_c1))),
                        rep(label_c2, length(mcols(granges_c2))))


  coldata <- data.frame(colnames(cts), condition = as.factor(condition_labels))

  dds <- DESeq2::DESeqDataSetFromMatrix(countData = cts,
                                colData = coldata,
                                design = ~ condition,
                                rowRanges = granges_c1[complete, ])


  if (estimate_size_factors == TRUE) {
    dds <- DESeq2::estimateSizeFactors(dds)
  }
  else {
    # Since files are scaled, we do not want to estimate size factors
    sizeFactors(dds) <- c(rep(1, ncol(cts)))
  }

  dds <- DESeq2::estimateDispersions(dds)
  dds <- DESeq2::nbinomWaldTest(dds)

  if (as_granges) {
    result <- DESeq2::results(dds, format = "GRanges",alpha = 0.01)
    if (!is.null(names_values)) {
      result$name <- names_values[complete]
    }

  }
  else {
    # result <- results(dds, format="DataFrame")
    result <- dds
  }

  result
}

get_nreads_columns <- function(df, length_factor = 100) {
  # Convert mean coverages to round integer read numbers
  cts <- as.matrix(df)
  cts <- as.matrix(cts[complete.cases(cts),])
  cts <- round(cts*length_factor)
  cts
}

BWs_Rloop <- paste0(bw_dir,list.files(bw_dir,pattern="Rloop_.+_R..bw"))
gencode <- "../genome/gencode.vm33.mm39.bed"
gencode <- "../genome/genes_lo_mid_hi_bi.mm39.bed"

BWs.WT <- BWs_Rloop[grep("WT",BWs_Rloop)]
BWs.FANCJ <- BWs_Rloop[grep("D1D6",BWs_Rloop)]
BWs.DHX36 <- BWs_Rloop[grep("P2D2",BWs_Rloop)]
BWs.DKO <- BWs_Rloop[grep("P3D4",BWs_Rloop)]
  
# Calculate here some loci or bins
cov.WT <- bw_loci(BWs.WT, loci = gencode)
cov.DHX36 <- bw_loci(BWs.DHX36, loci = gencode)
cov.FANCJ <- bw_loci(BWs.FANCJ, loci = gencode)
cov.DKO <- bw_loci(BWs.DKO, loci = gencode)

cov.WT$name <- paste0("peak_",1:length(cov.WT))
cov.DHX36$name <- paste0("peak_",1:length(cov.DHX36))
cov.FANCJ$name <- paste0("peak_",1:length(cov.FANCJ))
cov.DKO$name <- paste0("peak_",1:length(cov.DKO))

diff_DHX36 <- bw_granges_diff_analysis(cov.WT, cov.DHX36,
                                     "WT", "DHX36KO")

converting counts to integer mode
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates

diff_FANCJ <- bw_granges_diff_analysis(cov.WT, cov.FANCJ,
                                     "WT", "FANCJ")

converting counts to integer mode
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates

diff_DKO <- bw_granges_diff_analysis(cov.WT, cov.DKO,
                                     "WT", "DKO")

converting counts to integer mode
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates

# This takes care of low conts and things like this, but you can also use
# diff_results as is for the things below
lfc_DHX36 <- DESeq2::lfcShrink(diff_DHX36, coef = "condition_WT_vs_DHX36KO", type="apeglm")

using 'apeglm' for LFC shrinkage. If used in published research, please cite:
    Zhu, A., Ibrahim, J.G., Love, M.I. (2018) Heavy-tailed prior distributions for
    sequence count data: removing the noise and preserving large differences.
    Bioinformatics. https://doi.org/10.1093/bioinformatics/bty895

lfc_FANCJ <- DESeq2::lfcShrink(diff_FANCJ, coef = "condition_WT_vs_FANCJ", type="apeglm")

using 'apeglm' for LFC shrinkage. If used in published research, please cite:
    Zhu, A., Ibrahim, J.G., Love, M.I. (2018) Heavy-tailed prior distributions for
    sequence count data: removing the noise and preserving large differences.
    Bioinformatics. https://doi.org/10.1093/bioinformatics/bty895

lfc_DKO <- DESeq2::lfcShrink(diff_DKO, coef = "condition_WT_vs_DKO", type="apeglm")

using 'apeglm' for LFC shrinkage. If used in published research, please cite:
    Zhu, A., Ibrahim, J.G., Love, M.I. (2018) Heavy-tailed prior distributions for
    sequence count data: removing the noise and preserving large differences.
    Bioinformatics. https://doi.org/10.1093/bioinformatics/bty895

data_DHX36 <- plotMA(lfc_DHX36, returnData = T)
data_DHX36$lfc <- -data_DHX36$lfc
data_DHX36$mean <- log10(data_DHX36$mean)

data_FANCJ <- plotMA(lfc_FANCJ, returnData = T)
data_FANCJ$lfc <- -data_FANCJ$lfc
data_FANCJ$mean <- log10(data_FANCJ$mean)

data_DKO <- plotMA(lfc_DKO, returnData = T)
data_DKO$lfc <- -data_DKO$lfc
data_DKO$mean <- log10(data_DKO$mean)

ggscatter(data_DHX36,x ="mean",y="lfc",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[1])) + geom_hline(yintercept = 0, linetype="dashed", size=0.1) + coord_cartesian(ylim=c(-5,5))

plot_MA_DHX36 <- rasterize(last_plot(), layers='Point', dpi=600) 

data_DHX36$cov.WT <- rowMeans(as.data.frame(cov.WT)[,6:7])
data_DHX36$cov.DHX36 <- rowMeans(as.data.frame(cov.DHX36)[,6:7])
ggscatter(data_DHX36,x ="cov.WT",y="cov.DHX36",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[1])) + scale_x_continuous(trans="log",limits = c(0.5,400)) + scale_y_continuous(trans="log",limits = c(0.5,400)) + geom_abline(slope = 1, linetype="dashed", size=0.1)

plot_XY_DHX36 <- rasterize(last_plot(), layers='Point', dpi=600) 

data_DHX36$cov.WT <- rowMeans(as.data.frame(cov.WT)[,6:7])
data_DHX36$cov.DHX36 <- rowMeans(as.data.frame(cov.DHX36)[,6:7])
ggscatter(data_DHX36,x ="cov.WT",y="cov.DHX36",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[1])) + scale_x_continuous(limits = c(-10,50)) + scale_y_continuous(limits = c(-10,50)) + geom_abline(slope = 1, linetype="dashed", size=0.1)

plot_XY_DHX36_zoom <- rasterize(last_plot(), layers='Point', dpi=600) 

ggscatter(data_FANCJ,x ="mean",y="lfc",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[2])) + geom_hline(yintercept = 0, linetype="dashed", size=0.1) + coord_cartesian(ylim=c(-5,5))

plot_MA_FANCJ <- rasterize(last_plot(), layers='Point', dpi=600) 

data_FANCJ$cov.WT <- rowMeans(as.data.frame(cov.WT)[,6:7])
data_FANCJ$cov.FANCJ <- rowMeans(as.data.frame(cov.FANCJ)[,6:7])
ggscatter(data_FANCJ,x ="cov.WT",y="cov.FANCJ",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[2])) + geom_hline(yintercept = 0, linetype="dashed", size=0.1) + scale_x_continuous(trans="log",limits = c(0.5,400)) + scale_y_continuous(trans="log",limits = c(0.5,400)) + geom_abline(slope = 1, linetype="dashed", size=0.1)

plot_XY_FANCJ <- rasterize(last_plot(), layers='Point', dpi=600) 

ggscatter(data_DKO,x ="mean",y="lfc",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[3])) + geom_hline(yintercept = 0, linetype="dashed", size=0.1) + coord_cartesian(ylim=c(-5,5))

plot_MA_DKO <- rasterize(last_plot(), layers='Point', dpi=600) 

data_DKO$cov.WT <- rowMeans(as.data.frame(cov.WT)[,6:7])
data_DKO$cov.DKO <- rowMeans(as.data.frame(cov.DKO)[,6:7])
ggscatter(data_DKO,x ="cov.WT",y="cov.DKO",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[3])) + geom_hline(yintercept = 0, linetype="dashed", size=0.1) + scale_x_continuous(trans="log",limits = c(0.5,400)) + scale_y_continuous(trans="log",limits = c(0.5,400)) + geom_abline(slope = 1, linetype="dashed", size=0.1)

plot_XY_DKO <- rasterize(last_plot(), layers='Point', dpi=600) 

data_DKO$lfc_DHX36 <- data_DHX36$lfc
data_DKO$lfc_FANCJ <- data_FANCJ$lfc
ggscatter(data_DKO,x ="lfc_DHX36",y="lfc",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[3]))+ geom_hline(yintercept = 0, linetype="dashed", size=0.1) + geom_vline(xintercept = 0, linetype="dashed", size=0.1) + coord_cartesian(ylim=c(-5,5), xlim=c(-5,5))

plot_LFC_DKO_vs_DHX36 <- rasterize(last_plot(), layers='Point', dpi=600)

ggscatter(data_DKO,x ="lfc_FANCJ",y="lfc",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[3]))+ geom_hline(yintercept = 0, linetype="dashed", size=0.1) + geom_vline(xintercept = 0, linetype="dashed", size=0.1) + coord_cartesian(ylim=c(-5,5), xlim=c(-5,5))

plot_LFC_DKO_vs_FANCJ <- rasterize(last_plot(), layers='Point', dpi=600)

data_DHX36$lfc_DKO <- data_DKO$lfc
ggscatter(data_DHX36,x ="lfc",y="lfc_DKO",color="isDE",size = 0.8, alpha=0.5, palette = c("gray",mypal[1]))+ geom_hline(yintercept = 0, linetype="dashed", size=0.1) + geom_vline(xintercept = 0, linetype="dashed", size=0.1) + coord_cartesian(ylim=c(-5,5), xlim=c(-5,5))

cov <- cbind( as.data.frame(cov.WT)[,1:7], 
              as.data.frame(cov.DHX36)[,6:7], 
              as.data.frame(cov.FANCJ)[,6:7], 
              as.data.frame(cov.DKO)[,6:7])

colnames(cov) <- c(colnames(cov)[1:5],"WT1","WT2","DHX1","DHX2","FAN1","FAN2","DKO1","DKO2")

library(reshape2)
mdf <- melt(data.frame(name=cov.WT$name,cov[,6:13]))

Using name as id variables

mdf <- mdf[mdf$value<500,]
mdf$cond <- "WT"
mdf$cond[grep("DHX",mdf$variable)] <- "DHX36-/-"
mdf$cond[grep("FAN",mdf$variable)] <- "FANCJ-/-"
mdf$cond[grep("DKO",mdf$variable)] <- "DHX36-/-FANCJ-/-"
plot_viol_rep_all_peaks <- ggviolin(mdf, x="variable",y="value",fill="cond",palette = mypal, add="mean_sd") +
  coord_cartesian(ylim=c(0,10))
plot_viol_rep_all_peaks

plot_viol_rep_hi_genes <- ggviolin(mdf[mdf$name=="gene_hi",], x="variable",y="value",fill="cond",palette = mypal, add="mean_sd") +
  coord_cartesian(ylim=c(0,10))
plot_viol_rep_hi_genes

plot_viol_rep_biv_genes <- ggviolin(mdf[mdf$name=="gene_bi",], x="variable",y="value",fill="cond",palette = mypal, add="mean_sd") +
  coord_cartesian(ylim=c(0,10))
plot_viol_rep_biv_genes

plot_viol_rep_by_genes <- ggviolin(mdf, x="name",y="value",fill="cond",palette = mypal, add="mean_sd") +
  coord_cartesian(ylim=c(0,10))
plot_viol_rep_by_genes